A statistical approach for carrier particle optimisation – Design of Experiments & Bayesian Optimisation
Melvin Wostry & Regina Scherließ
Department of Pharmaceutics and Biopharmaceutics, Kiel University, Gutenbergstraße 76, 24118 Kiel, Germany
Summary
Systematic modifications of a geometry for the carrier particle in interactive powder mixtures were evaluated for their effects on the drug detachment during a simulated inhalation. The inhalation was mimicked with an established in-silico model in which an adhesive unit of carrier and drug was created and via different types of collisions, the detachment of drug was induced. Using a parametric design approach, the carrier geometry was modified step wise. For this, four geometry parameters were defined that were to be tested within given limitations. With two different statistical tools, influential geometry features were found and optimised carrier particle were created. With the Design of Experiments, it was found that surfaces spikes were the most influential parameter, with larger spikes resulting in a better drug detachment. Surface indentations on the other hand were found to be detrimental for the drug dispersion. With the Bayesian Optimisation, a different tool for an iteration-based approximation to the optimal carrier geometry was tested. An automation of the whole iteration process was achieved with several Python scripts. A total of 502 individual simulations were created, calculated and analysed, creating the output parameter used in the Bayesian Optimisation for suggesting the next carrier geometry. Both statistical models, Design of Experiments and Bayesian Optimisation, lead to a better understanding of crucial geometry features and resulted in a carrier particle geometry with better performance than the initial carrier geometries, thus the optimisation was successful.